The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external CSV, TSV, and AnnData files provided by the user, which represents a standard surface for processing untrusted data.
Ingestion points: Data ingestion is handled in scripts/run_deseq2_analysis.py via pandas.read_csv for both genomic count matrices and sample metadata.
Boundary markers: The skill documentation in SKILL.md and references/workflow_guide.md contains explicit security advice, warning users to avoid pickle files from untrusted sources and to prefer structured formats like H5AD.
Capability inventory: The skill uses Bash to execute Python scripts via the uv package manager and requires Write access to export analysis results (CSV) and plots (PNG).
Sanitization: The provided analysis script includes a validation routine in load_and_validate_data that checks for negative values in count matrices and verifies sample index alignment.

pydeseq2